Yiqing Xu

National University of Singapore
Incoming Postdoctoral Scholar, Stanford University
yiqingx[at]stanford[dot]edu

prof_pic.jpg

I’m Yiqing Xu, a CS Ph.D. candidate at the National University of Singapore, advised by Prof. David Hsu, and an incoming postdoctoral scholar at Stanford University, advised by Prof. Jiajun Wu. I obtained double degrees in Computer Science and Applied Mathematics from NUS.

Previously, I was a visiting Ph.D. student at MIT CSAIL, advised by Prof. Leslie Kaelbling and Prof. Tomás Lozano-Pérez, from September 2023 to February 2024. From November 2025 to May 2026, I was a robotics intern at the Allen Institute for AI, advised by Prof. Dieter Fox, where I worked on translating human intent into robot goal specifications for tabletop manipulation tasks.

research highlights

My research focuses on translating human objectives into representations that robots can optimize. I study how robots can understand goals expressed through intuitive but often ambiguous human inputs, such as language, sketches, demonstrations, and multi-modal interaction. Across my work, I design intermediate representations, including compositional, relational, and hierarchical structures, that bridge the gap between human intent and robot reasoning.

In my recent works, “Set It Up” (IJRR) and “Stack It Up” (CoRL), I explore how robots can act on goals conveyed through language commands and freehand sketches. Both systems map human inputs into abstract relational representations and then ground them into feasible physical configurations using compositional generative models. These approaches preserve task structure, support generalization, and enable robots to learn from limited data by reusing local relational priors.

More broadly, I am interested in goal specification as an interface problem: how humans communicate what they want, how robots represent those objectives internally, and how those representations can be optimized into action. My current and future work extends this direction toward flexible skill chaining, multi-step reasoning, and interactive multi-modal goal specification. This includes inferring symbolic task skeletons and modular reward functions from mixed human inputs, as well as enabling robots to resolve ambiguity through dialogue, gaze, motion, and active inference.

The long-term goal of my research is to make robots more capable of understanding and acting on human intent in ways that are expressive, adaptable, and aligned with how people naturally communicate goals.

If you’d like to chat more, feel free to email me!

selected publications

  1. RHIRL.png
    Receding Horizon Inverse Reinforcement Learning
    Yiqing Xu, Wei Gao, and David Hsu
    In Advances in Neural Information Processing Systems (NeurIPS), 2022
    arXiv
  2. LLMReward.png
    Learning Reward for Physical Skills using Large Language Model
    Yuwei Zeng, and Yiqing Xu
    In Conference on Robot Learning (CoRL), LangRob workshop, 2023
    arXiv
  3. TidyItUp.png
    "Tidy Up the Table": Grounding Common-sense Objective for Tabletop Object Rearrangement
    Yiqing Xu, and David Hsu
    In Robotics: Science and Systems (RSS), LangRob workshop, 2023
    arXiv
  4. Effective_IRL.png
    On the Effective Horizon of Inverse Reinforcement Learning
    Yiqing Xu, Finale Doshi-Velez, and David Hsu
    In Proceedings of the International Conference on Autonomous Agents and Multiagent Systems (AAMAS), 2025
    arXiv
  5. SetItUp.png
    "Set It Up!": Functional Object Arrangement with Compositional Generative Models (Conference Version)
    Yiqing Xu, Jiayuan Mao, Yilun Du, and 3 more authors
    In Robotics: Science and Systems (RSS), 2024
    arXiv
  6. setitup_ijrr.png
    “Set It Up": Functional Object Arrangement with Compositional Generative Models (Journal Version)
    Yiqing Xu, Jiayuan Mao, Linfeng Li, and 4 more authors
    The International Journal of Robotics Research (IJRR), 2025
    arXiv
  7. stackitup.png
    “Stack It Up!": 3D Stable Structure Generation from 2D Hand-drawn Sketch
    Yiqing Xu, Linfeng Li, Cunjun Yu, and 1 more author
    In Proceedings of the Conference on Robot Learning (CoRL), 2025
    arXiv